What is MCP (Model Context Protocol)?

On this page

Definition
Simple explanation
Why it matters
How it works
Real examples
Advantages
Limitations
Misunderstandings
Related terms
FAQs
TTML perspective
Conclusion

Definition

The Model Context Protocol is an open standard for connecting AI models to external tools, data and services through a shared JSON-RPC interface.

Simple explanation

Before MCP, every AI product invented its own way to connect to Slack, Google Drive, Notion or a database. MCP is the shared cable — one integration works across many AI clients.

Why it matters

MCP is what will make agents genuinely useful at work. It turns 'AI that can chat' into 'AI that can act inside your stack' without one-off code for every combination.

How it works

1
Server
A tool provider exposes an MCP server describing its capabilities.
2
Client
The AI application connects as a client.
3
Discover
The client asks what tools are available.
4
Call
The model chooses tools and calls them via MCP.

Real examples

Products named for illustration only. Inclusion is not an endorsement.

Claude Desktop
One of the first clients to ship MCP support.
Cursor
Uses MCP to connect editors to external services.
Public MCP servers
Growing ecosystem for GitHub, Slack, Postgres and more.

Advantages

Standardised — one integration, many clients.
Open — not owned by a single vendor.
Secure by design — capabilities are declared.

Limitations

Early — best practices are still emerging.
Ecosystem is uneven; not every tool has a server yet.
Governance and audit tooling is immature.

Common misunderstandings

Claim
MCP is an AI model.
Reality
It is a protocol — the wire between models and tools.
Claim
MCP is only for developers.
Reality
End-users benefit; they just see 'connect Slack' work everywhere.

Frequently asked questions

Who created MCP?

Anthropic released it as an open standard; adoption is now cross-vendor.

Is MCP replacing APIs?

No — it sits on top of APIs to make them AI-friendly.

Do I need MCP to build agents?

No, but it dramatically reduces integration work.

The Tool Money Lab perspective

MCP is the quiet infrastructure story of the AI decade. Protocols always look boring until every serious product depends on them; MCP is on that trajectory.

Conclusion

MCP standardises how AI reaches the rest of your software. Expect it to become as foundational to AI as HTTP is to the web.